On Sat, 31 Mar 2018, 03:07 Chris Murphy, <lists@xxxxxxxxxxxxxxxxx> wrote:
On Fri, Mar 30, 2018 at 8:15 AM, Gerald Henriksen <ghenriks@xxxxxxxxx> wrote:
> Consult the relevant experts, and based on their recommendations
> mandate a base set of fonts that provide a quality first experience
> with Fedora for everyone regardless of where they live and what
> language they read/write. Making compromises to save 400M on a
> distribution that will need 20+G over year doesn't seem very wise.
Compression of small data like this might benefit from tuning. Zstd
offers a training mode to produce a dictionary used for compression,
offering upwards of 3x more compression. Speed of compression and
decompression is also improved. Zstd even without a dictionary
outperforms xz, in particular decompression speeds. But at the sheer
volume of xz (or bzip2?) compression for RPMs and LiveOS images in
koji, I'd bet the compression time savings would be significant,
without dictionary. And if training is possible, the advantage in
compression ratio and performance would be huge.
The compression used in rpms is mostly irrelevant in these cases because when the rpm is installed on disk it's not compressed.
You also have the problem of availability of tools to handle the compression, xz is a widely used format which has advantages too. In terms of training I suspect given the wide range of data that gets jammed into rpms you'd find the overall improvement is minimal.
The Zstd patches for squashfs-tools don't appear to be upstream still
for whatever reason, but patches are available. The kernel code for
squashfs and Btrfs has been there for a while. So implementing this
for LiveOS images is probably a good deal easier than whatever RPM and
koji would have to learn to make it happen for all rpms.
Live images are but one of the artifacts that we distribute, we also have cloud, container, vagrant and the arm images all of which could be used on top of other OSes which would need wide support for the format. Btrfs is also not a default in any of the delivered artefacts.
Peter
--
Chris Murphy
_______________________________________________
devel mailing list -- devel@xxxxxxxxxxxxxxxxxxxxxxx
To unsubscribe send an email to devel-leave@xxxxxxxxxxxxxxxxxxxxxxx
_______________________________________________ devel mailing list -- devel@xxxxxxxxxxxxxxxxxxxxxxx To unsubscribe send an email to devel-leave@xxxxxxxxxxxxxxxxxxxxxxx
